{"slug": "nvidia-cosmos-opens-new-tools-to-train-evaluate-and-deploy-world-models", "title": "NVIDIA Cosmos Opens New Tools to Train, Evaluate and Deploy World Models", "summary": "NVIDIA expanded its Cosmos platform with new training tools, simulation checkpoints and five cloud-native Cosmos Data Factory blueprints for building world models for physical AI, available through GitHub, Microsoft Azure, Nebius and CoreWeave. NVIDIA said the Cosmos platform ranks in the top five on six different open models leaderboards, and Porsche is using Cosmos Data Factory via Voxel51 on Microsoft Azure to build a shared autonomous-vehicle data flywheel across Volkswagen Group brands, which is expected to cut repeated physical test drives by up to 90% and shorten development cycle times by up to 85%. ANYbotics generated over 1,500 synthetic video clips with the blueprints, validating outputs with Encord's platform, to train industrial-inspection robots to recognize liquid spills and other changes between facility visits.", "body_md": "# NVIDIA Cosmos Opens New Tools to Train, Evaluate and Deploy World Models\n\n[NVIDIA Blog](https://blogs.nvidia.com)\n\nTo help developers rapidly build world models, which help physical AI systems understand and simulate the real world, NVIDIA has expanded the Cosmos platform with new training tools, simulation checkpoints and Cosmos Data Factory blueprints. Ranking top five on six different open models leaderboards, the NVIDIA Cosmos platform gives world model builders the frontier open […]\n\nTo help developers rapidly build [world models](https://www.nvidia.com/en-us/glossary/world-models/), which help [physical AI](https://www.nvidia.com/en-us/glossary/generative-physical-ai/) systems understand and simulate the real world, NVIDIA has expanded the [Cosmos platform](https://www.nvidia.com/en-us/ai/cosmos/) with new training tools, simulation checkpoints and Cosmos Data Factory blueprints. \n\nRanking top five on six different open models leaderboards, the NVIDIA Cosmos platform gives world model builders the frontier open models — as well as data, training and optimization tools and recipes — to scale training data, adapt models, test behavior and optimize for performance before deployment.\n\n## **Build Better Training Data**\n\nBuilding world models begins with multimodal, multisensory data.\n\n[NVIDIA Cosmos Data Factory](https://github.com/NVIDIA/cosmos-data-factory) brings together the tools and skills for labeling, curating, augmenting and evaluating data at scale for training physical AI. Cosmos Curator, a GPU-accelerated solution for curating video training datasets, has now added support for multimodal multisensory inputs and processing blocks.\n\nCosmos Data Factory now offers the following five cloud-native blueprints and is available through [GitHub](https://github.com/NVIDIA/cosmos-data-factory), [Microsoft Azure](https://www.microsoft.com/en-us/microsoft-cloud/blog/mobility/automotive/2026/10/21/advancing-autonomous-vehicle-development-with-microsoft-nvidia-porsche-and-voxel51/), [Nebius](https://nebius.com/blog/posts/nebius-expands-physical-ai-workbench-agentic-workflows-nvidia-cosmos%E2%81%A0%EF%BF%BD) and [CoreWeave](https://coreweave.com/news/coreweave-unveils-coreweave-physical-ai-launchables-taking-the-setup-work-out-of-robotics-and-av-development):\n\n- **Autonomous Vehicle Dataset Generation:** Curates and labels driving data, then adds synthetic examples of hard-to-capture scenarios to expand training coverage.\n- **Image Attribute Augmentation:** Varies object appearance to improve image search.\n- **Event Video Generation:** Turns still images into synthetic videos of rare or unsafe events.\n- **Defect Image Generation:** Generates printed circuit board, metal and glass defect images for inspection models.\n- **Video [Data Augmentation](https://www.machinebrief.com/glossary/data-augmentation):** Changes weather and lighting in existing footage while preserving scene structure.\n\n*Cosmos Data Factory accelerates data curation, processing and [synthetic data](https://www.machinebrief.com/glossary/synthetic-data) generation with agentic skills.*\n\nCompanies are already putting Cosmos Data Factory blueprints to work to speed their physical AI development workflows.\n\nPorsche is leading the development of a shared AV data flywheel across Volkswagen Group brands, using Cosmos Data Factory through Voxel51 on Microsoft Azure. This is expected to reduce repeated physical test drives by up to 90% and shorten development cycle times by up to 85%.\n\n*Porsche searching and augmenting autonomous vehicle data footage in Voxel51, powered by Cosmos Data Factory.*\n\nANYbotics generated over 1,500 synthetic video clips with Cosmos Data Factory blueprints, curating real-world data and validating outputs with [Encord’s](https://encord.com/data-labs/nvidia-anybotics-synthetic-data/) platform, to train its industrial-inspection robots to recognize liquid spills and other changes between facility visits.\n\n*ANYbotic**s* *generating anomaly clips with* *Encord* *powered by NVIDIA Cosmos Data Factory.*\n\n[Ultralytics](https://www.ultralytics.com/blog/nvidia-cosmos-yolo-synthetic-defect-detection?utm_source=nvidia&utm_medium=referra), a [computer vision](https://www.machinebrief.com/glossary/computer-vision) company, retrained its YOLO model with real images and targeted synthetic defects generated using Cosmos Data Factory, improving defect localization and recall.\n\n## **Train, Adapt and Optimize Models**\n\nWith training data prepared, developers may train their own architecture or start with a pretrained model.\n\nAt GTC Berlin, NVIDIA extended [Cosmos Framework](https://github.com/NVIDIA/cosmos-framework) training, optimization and deployment tools and recipes as agent skills to more open models, along with Cosmos 3, giving world model builders a faster path from experimentation to deployment.\n\nDevelopers and agents can now adapt open models like GR00T, Alpamayo, QWEN, Pi0.5 and other compatible Hugging Face models through fine-tuning and [reinforcement learning](https://www.machinebrief.com/glossary/reinforcement-learning), optimize the model with [distillation](https://www.machinebrief.com/glossary/distillation) and quantization, and deploy with high-performing inference.\n\nTo develop a world model specialized for egocentric manipulation, robotics research lab MicroAGI post-trained Cosmos 3 on videos of humans manipulating objects and distilled its 32-billion-parameter reasoner down to an 8-billion-parameter Nano model with no loss in accuracy.\n\nDevelopers can label data and post-train Cosmos 3 on [Roboflow](http://google.com/url?q=http://blog.roboflow.com/roboflow-joins-nvidia-cosmos-coalition&sa=D&source=docs&ust=1792534578709344&usg=AOvVaw3wuHbhvvFjmEwKpqE36ZSs) for their specific environments and tasks, then deploy those models to RoboFlow hosted or on-premises NVIDIA GPUs to bring video reasoning into factories and warehouses.\n\n## **Test Policies and Run Models Faster** \n\nAlongside these tools, NVIDIA continues advancing foundational technology through its work on Cosmos models.\n\nNew [Cosmos 3 Nano model variants](https://huggingface.co/collections/nvidia/cosmos3-examples) for simulation enable developers to test autonomous policies in repeatable, hard-to-capture, multiview multisensor scenarios. Accompanying code allows developers to adapt the models to train and evaluate their own policies for autonomous vehicles and bimanual robots in interactive simulations before real-world testing.\n\nDevelopers can deploy core Cosmos models using updated [NVIDIA Cosmos 3 NIM](https://catalog.ngc.nvidia.com/orgs/nim/nvidia/containers/cosmos3/-) microservices, which deliver up to 2x faster inference with no quality loss in validated configurations, while continuous video analysis enables sub-second alerts. \n\nFor vision AI developers, the new [NVIDIA Metropolis Blueprint for video search and summarization (VSS) 3.3](https://build.nvidia.com/nvidia/video-search-and-summarization), built on Cosmos NIM microservices, turns one prompt into an efficient video AI application and helps teams monitor 40% more video streams with the same GPU. \n\nADNOC uses Fogsphere’s MistIQ agentic platform, built with Cosmos and the [NVIDIA Metropolis VSS Blueprint](https://build.nvidia.com/nvidia/video-search-and-summarization), to detect safety hazards in real time across 65 sites. For rare events, [Fogsphere](https://fogsphere.com/fogsphere-scales-ai-agents-at-adnoc/) also uses the Cosmos Data Factory to generate synthetic training data, improving detection without needing to physically stage dangerous situations.\n\n*The Cosmos platform is fully modular and open. Developers can use* *Cosmos Data Factory* *to generate training datasets,* *Cosmos Framework* *to train and adapt world models, then deploy supported* *Cosmos models through NIM microservices**.*\n\nGet AI news in your inbox\n\nDaily digest of what matters in AI.\n\n## Key Terms Explained\n\n[Computer Vision](https://www.machinebrief.com/glossary/computer-vision)\n\nThe field of AI focused on enabling machines to interpret and understand visual information from images and video.\n\n[Data Augmentation](https://www.machinebrief.com/glossary/data-augmentation)\n\nTechniques for artificially expanding training datasets by creating modified versions of existing data.\n\n[Distillation](https://www.machinebrief.com/glossary/distillation)\n\nA technique where a smaller 'student' model learns to mimic a larger 'teacher' model.\n\n[Fine-Tuning](https://www.machinebrief.com/glossary/fine-tuning)\n\nThe process of taking a pre-trained model and continuing to train it on a smaller, specific dataset to adapt it for a particular task or domain.", "url": "https://wpnews.pro/news/nvidia-cosmos-opens-new-tools-to-train-evaluate-and-deploy-world-models", "canonical_source": "https://www.machinebrief.com/news/nvidia-cosmos-opens-new-tools-to-train-evaluate-and-deploy-w-9cit", "published_at": "2026-10-22 05:00:27+00:00", "updated_at": "2026-10-22 05:42:04.690684+00:00", "lang": "en", "topics": ["artificial-intelligence", "robotics", "autonomous-vehicles", "ai-infrastructure", "generative-ai"], "entities": ["NVIDIA", "NVIDIA Cosmos", "Cosmos Data Factory", "Porsche", "Volkswagen Group", "Voxel51", "Microsoft Azure", "ANYbotics"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/nvidia-cosmos-opens-new-tools-to-train-evaluate-and-deploy-world-models", "markdown": "https://wpnews.pro/news/nvidia-cosmos-opens-new-tools-to-train-evaluate-and-deploy-world-models.md", "text": "https://wpnews.pro/news/nvidia-cosmos-opens-new-tools-to-train-evaluate-and-deploy-world-models.txt", "jsonld": "https://wpnews.pro/news/nvidia-cosmos-opens-new-tools-to-train-evaluate-and-deploy-world-models.jsonld"}}